Originally posted by: ouaigooo

Hi

Post-processing sometimes crashes when accessing "dexpr" after having called convertAllIntVars().

Exception from IBM ILOG Concert: convertAllIntVars() is not supported for this dexpr element.

What are the rules to respect in order to avoid this ?

Thanks for your help !

Hi,

let me give you an example derived from the convert example that works:

dvar int x in 0..10;

dexpr float y=2*x;
minimize x;
subject to {
  ct :   
    x >= 1/2;
}

execute
{
writeln("x=",x);

writeln("y=",y);
}

main {
  var status = 0;
  thisOplModel.generate();
  if (cplex.solve()) {
    writeln("Integer Model");   
    writeln("OBJECTIVE: ",cplex.getObjValue());  
    thisOplModel.postProcess() ;
    if (cplex.getObjValue() != 1) {
      status = -1;
    }
  }

  thisOplModel.convertAllIntVars();
  if (cplex.solve()) {
    writeln("Relaxed Model");   
    writeln("OBJECTIVE: ",cplex.getObjValue());  
    //thisOplModel.postProcess() ;  // indeed uncommenting this breaks
    if (cplex.getObjValue() != 0.5) {
      status = -1;
    }
  }
   
  thisOplModel.unconvertAllIntVars();
  if (cplex.solve()) {
    writeln("Unrelaxed Model");   
    writeln("OBJECTIVE: ",cplex.getObjValue());
    thisOplModel.postProcess() ;
    if (cplex.getObjValue() != 1) {
      status = -1;
    }
  }
  status;
}

regards

Originally posted by: ouaigooo

Thanks, but it doesn't help that much.

I could indeed expect that there exist examples where it works.

But the question is: why it doesn't work in my case ?

With the subsidary question: when does it work and when not ?

Here is more info about my code:

It crashes in an execute block (post-processing):

for (var t in OPTSTEPS) f.writeln(t.start + ";" + t.stop + ";" + TIME_STAMP[t] + ";" + power_el[i][t]);

In the model file, we have:

dvar float boiler_Pel[BOILERS,SCENARII,BSTEPS] in -10..10;

dvar float+ machine_power_var[MACHINES diff BOILERS][SCENARII][OPTSTEPS];
dexpr float machine_power[x in MACHINES][i in SCENARII][t in OPTSTEPS] =
 ((x in BOILERS) ? boiler_Pel[x,i,Bstep[t]] : machine_power_var[x,i,t]);

dexpr float power_el[i in SCENARII][o in OPTSTEPS] = 0
    + sum(x in RESIDUAL_POWER: (o.start < x.stop) && (o.stop > x.start)) x.value * (minl(o.stop, x.stop) - maxl(o.start, x.start)) / (o.stop - o.start)
    + sum(x in MACHINES) item(MACHINE_SIGN,<x>).value * machine_power[x,i,o]
;

RESIDUAL_POWER is pure data.

Any guess why it crashes and how I could adapt ?

Hi,

the issue is with dexpr float that relates to dvar int so a workaround is to use dvar float vars in order to help.

Let me give you an example that works fine:

dvar int x in 0..10;
dvar float fx in 0..10;

dexpr float y=2*fx;
minimize x;
subject to {
fx==x;

  ct :   
    x >= 1/2;
}

execute
{
writeln("x=",x);

writeln("y=",y);
}

main {
  var status = 0;
  thisOplModel.generate();
  if (cplex.solve()) {
    writeln("Integer Model");   
    writeln("OBJECTIVE: ",cplex.getObjValue());  
    thisOplModel.postProcess() ;
    if (cplex.getObjValue() != 1) {
      status = -1;
    }
  }

  thisOplModel.convertAllIntVars();
  if (cplex.solve()) {
    writeln("Relaxed Model");   
    writeln("OBJECTIVE: ",cplex.getObjValue());  
    thisOplModel.postProcess() ;
    if (cplex.getObjValue() != 0.5) {
      status = -1;
    }
  }
   
  thisOplModel.unconvertAllIntVars();
  if (cplex.solve()) {
    writeln("Unrelaxed Model");   
    writeln("OBJECTIVE: ",cplex.getObjValue());
    thisOplModel.postProcess() ;
    if (cplex.getObjValue() != 1) {
      status = -1;
    }
  }
  status;
}

which gives

Integer Model
OBJECTIVE: 1
x=1
y=2
Relaxed Model
OBJECTIVE: 0.5
x=0.5
y=1
Unrelaxed Model
OBJECTIVE: 1
x=1
y=2

regards

Originally posted by: ouaigooo

Hi Alex,

You said:

the issue is with dexpr float that relates to dvar int so a workaround is to use dvar float vars in order to help.

I'm afraid we don't speak about the same.

By "my case", I mean my problem, not your example.

There is no dvar int in my code.

Can you help to answer the 2 questions:

But the question is: why it doesn't work in my case ?

With the subsidary question: when does it work and when not ?

or shall I report (again) to CPLEX support ?

Regards

Hi,

since you wrote that

convertAllIntVars().

did not work I thought some dvar int were involved.

About

        But the question is: why it doesn't work in my case ?

        With the subsidary question: when does it work and when not ?

It is quite hard to answer without a full example that does not work.

What you could try as a workaround is to replace convertAllIntVars() by some other way of linear relaxation.

Any way, the most efficient way to get an answer is always to open a PMR though support

regards

Originally posted by: ouaigooo

I'll open a PMR. Thanks for your help anyway !